What Can I Expect From the Future of Machine Learning?

Machine learning has come a long way since the 1952 introduction of the first computer learning program.

In 1980, data scientists created Explanation Based Learning, where a computer could create a rule and discard irrelevant data. Nearly three decades later, a Google AI beat a professional player in a 2016 game of Go, a complex Chinese board game.

Now, in 2022, there’s no indication that these developments will slow anytime soon. In fact, the market is expected to grow to $20 billion by 2025.

Here are three machine learning trends you can expect to see, and take advantage of, in the coming years. 

3 Rising Trends in Machine Learning Design

1. A Reduced Need for Large Amounts of Data

Machine learning used to require large datasets to find desired patterns. Now, however, it is more important that organizations have the right data. Using the right data means prioritizing its quality and relevance over the number of points in a dataset.

Organizations that previously lacked sufficient data can now build models using the right data (supplemented by third-party sources, when needed). The need for less data also increases AI’s overall accessibility of AI—meaning that organizations with lower budgets and smaller teams can also reap the benefits of AI.

2. Increased Use of Multimodal Data 

In order for a machine learning model to truly understand what’s going on around us, it has to analyze multimodal data. Just as humans use five senses to process their surroundings, a single machine learning model will learn to understand spreadsheets, images, text, audio, and other mediums. 

3. Development, Use, and Adaptation of Foundational Models

We’re also starting to see AI projects benefiting from transfer learning, where a foundational machine learning model trained on a large amount of data is reused as a “starting point” to train another model for another task. Models like these can improve the efficiency of AI design by eliminating the need to start over on every project. 

Limitations of Using Foundational Machine Learning Models

While foundational machine learning models lower the barriers to developing and piloting AI, they also pose several limitations to organizations. Below are a few.

Unpredictable Third-Party Platforms

Google, Meta, OpenAI, and Microsoft all offer their own foundational models for organizations to use. While these platforms allow a range of companies access to ML capabilities, there is a marked lack of control and unpredictability when it comes to using them. Organizations are at the mercy of the third party for model upgrades and changes, and may even fall victim to the complete disposal of the model.

Restricted Transparency of Training Data 

Because foundational models are owned by third-party entities and trained on mass amounts of data, it’s not possible to monitor the quality of this training data. 

Moreover, a machine learning model continues to learn on its own after it is fed initial input data—which means it may draw conclusions based on what it learns as true, even if it’s inaccurate. This, combined with the lack of control over foundational models, makes it challenging to monitor for bias and other risk factors. 

Inconsistent Measurement of Bias 

Bias is measured subjectively and as Google’s Chief Decision Scientist, Cassie Kozyrkov mentions, its definition varies depending on the context it’s used in. In data collection, bias is when your sample isn’t representative of your population of interest. In AI, algorithmic bias occurs when the model reflects the implicit values of the humans who created it.

Humans with different backgrounds and experiences will detect harmful and discriminatory bias at different levels. This inconsistent monitoring is only accelerated when using a foundational model—one that was developed by yet another group of humans with their own biases.   

Machine learning is constantly evolving. And although we can’t predict the future with absolute certainty, we can expect these trends to gain momentum soon. 

Find More Answers to Your AI Questions 

This post is part of a series that features answers to top AI questions, from our data science experts. Learn more about the future of machine learning—and discover answers to other readers’ top AI questions—by subscribing to our Voices of Trusted AI monthly digest. It’s a once-per-month email that contains helpful trusted AI resources, reputable information and actionable tips straight from data science experts.

Contributor: Cal Al-Dhubaib is the CEO and AI Strategist at Pandata.